A surrogate endpoint is a laboratory measurement, imaging finding, or other biomarker used in a clinical trial as a substitute for a direct measure of how a patient feels, functions, or survives. Tumor shrinkage standing in for overall survival, LDL cholesterol standing in for cardiovascular events, and CD4 count standing in for HIV-related mortality are all surrogate endpoints in this sense. They exist because the true clinical outcome a trial ultimately cares about — survival, irreversible morbidity, symptomatic benefit — can take years or a very large sample size to observe directly. A validated surrogate lets a sponsor and a regulator reach a reliable conclusion faster, on fewer patients, which matters most for serious and life-threatening conditions where a years-long confirmatory trial delays access to a genuinely effective treatment.
The catch is that a surrogate is only useful if it actually predicts the clinical outcome it is standing in for — and the history of drug development includes real, costly cases where it did not. This guide covers what makes an endpoint a surrogate rather than a biomarker used for other purposes, the statistical and biological framework used to validate one, and how the FDA’s accelerated approval pathway relies on surrogate and intermediate clinical endpoints of varying evidentiary strength.
Surrogate Endpoint vs. Biomarker vs. Clinical Outcome Assessment
These three terms are related but not interchangeable, and mixing them up is a common source of confusion in protocol design:
- Biomarker — any defined, objectively measured characteristic used as an indicator of a normal biological process, a pathogenic process, or a response to an exposure or intervention (the BEST glossary’s definition, jointly maintained by the FDA and NIH). A biomarker is the broadest category; it becomes a surrogate endpoint only once it is validated to predict a specific clinical outcome and used for that purpose in a trial.
- Surrogate endpoint — a biomarker or intermediate measure used as a substitute for how a patient feels, functions, or survives, specifically to support a conclusion about clinical benefit.
- Clinical Outcome Assessment (COA) — a direct measure of how a patient feels or functions (patient-reported, clinician-reported, observer-reported, or performance-outcome based). A COA is not a surrogate; by definition it measures the thing itself rather than substituting for it.
Most trials that rely on randomization to control bias still need this endpoint hierarchy decided at the protocol-design stage, well before enrollment — see Clinical Trial Phases for where endpoint selection fits into the overall development timeline, since surrogate endpoints are used differently across phases (exploratory in early phase, confirmatory-adjacent late phase).
Why Trials Use Surrogate Endpoints at All
Three practical pressures drive surrogate endpoint use, and a research administrator building or reviewing a protocol budget should recognize all three because each has cost and timeline implications:
- Trial duration. Waiting for a hard clinical outcome (death, irreversible disability, disease progression) can take years; a biomarker that changes earlier lets a trial reach a conclusion sooner.
- Sample size and feasibility. Rare or slow-accumulating clinical events require large cohorts to detect a treatment effect with adequate statistical power. A more common, earlier-changing surrogate can need a fraction of the enrollment.
- Ethical urgency. In serious or life-threatening conditions with no adequate existing therapy, a validated or reasonably likely surrogate can support earlier access to a drug that appears to work, with confirmatory evidence gathered afterward.
None of this eliminates the need for validation — it is precisely because a surrogate is used to shortcut the direct evidence that the framework for validating one has to be rigorous.
The Validation Framework
Biological Plausibility
The starting requirement is a documented, mechanistic reason to expect the surrogate lies on the causal pathway between the intervention and the clinical outcome — not merely that the two happen to move together. A surrogate with strong biological plausibility (for example, viral load in HIV treatment, which sits directly on the causal pathway to HIV-related morbidity) is a much stronger validation candidate than one whose relationship to the outcome is only observational. Plausibility alone is necessary but not sufficient; it establishes the hypothesis a formal validation exercise then has to test statistically.
The Prentice Criteria
The most-cited formal statistical definition of a valid surrogate comes from Ross Prentice’s 1989 framework, which requires that the treatment’s full effect on the true clinical outcome be captured — fully mediated — through its effect on the surrogate. In practice this is a demanding, often unmeetable bar: few biological pathways are so completely captured by a single measurable marker that 100% of a treatment’s effect on survival or morbidity runs through it. The Prentice criteria remain the conceptual reference point for what “full validation” would mean, but they are widely recognized in the biostatistics literature as too strict to apply directly to most real surrogate-validation decisions.
The Meta-Analytic (Trial-Level) Validation Approach
Because strict Prentice-style mediation is rarely demonstrable, the dominant practical validation method today asks a more tractable question at the level of multiple trials rather than a single patient: across a body of randomized trials testing different interventions in the same disease, does a treatment-induced change in the surrogate reliably predict a treatment-induced change in the true clinical outcome? This trial-level (meta-analytic) approach evaluates the correlation of treatment effects across trials, not just the correlation of the surrogate and outcome within one trial — a distinction that matters because a surrogate can track an outcome well within a single trial’s patient population while still failing to predict how a *different* mechanism of action will affect that outcome. This is the core reason a surrogate validated for one drug class cannot automatically be assumed valid for a different class acting through a different pathway.
FDA’s Two-Tier Framework: Validated vs. Reasonably Likely
The FDA’s own resources on this topic (BEST glossary, jointly maintained with NIH) draw a specific two-tier distinction that determines how much regulatory weight an endpoint can carry:
- Validated surrogate endpoint — supported by a clear mechanistic rationale and strong clinical evidence, sufficient on its own to support traditional approval without requiring further outcome data after approval.
- Reasonably likely surrogate endpoint — supported by strong mechanistic and/or epidemiologic rationale, but without clinical evidence sufficient to reach the “validated” bar. This tier can support the accelerated approval pathway specifically, subject to a required postmarketing confirmatory trial.
FDA maintains a running Table of Surrogate Endpoints That Were the Basis of Drug Approval or Licensure, and a dedicated Surrogate Endpoint Resources for Drug and Biologic Development hub, as primary references for which endpoints currently sit in which tier for a given disease area.
The Accelerated Approval Pathway’s Reliance on Surrogate Endpoints
Accelerated approval is codified at 21 CFR Part 314 Subpart H (drugs, §§314.500–314.560) and 21 CFR Part 601 Subpart E (biologics, §§601.40–601.46). It allows FDA to grant marketing approval based on an effect on a surrogate endpoint, or an intermediate clinical endpoint, that is reasonably likely to predict clinical benefit — rather than requiring the direct clinical-benefit evidence a traditional approval needs before approval is granted. Two eligibility conditions gate the pathway: the product must treat a serious or life-threatening condition, and it must offer a meaningful therapeutic advantage over available therapy. Under 21 CFR 314.510 / 601.41, FDA may weigh “epidemiologic, therapeutic, pathophysiologic, or other evidence” in deciding whether an endpoint clears the reasonably-likely bar — a deliberately broader standard of evidence than full Prentice-style validation would require.
The regulatory trade-off is explicit, not implicit: approval under this pathway is conditioned on the sponsor completing a postmarketing confirmatory trial to verify the anticipated clinical benefit actually materializes. If the confirmatory trial fails to verify benefit, or is not completed with due diligence, FDA can withdraw the approval through an expedited process. For the operational detail of how confirmatory-trial and other postmarketing obligations are tracked and reported once an accelerated approval is granted, see the CASRAI dictionary’s treatment of Good Pharmacovigilance Practices (GVP) and the Pharmacovigilance in Clinical Research guide, which cover the adverse-event side of the same postmarketing period.
Why Weak Surrogates Are a Real, Documented Risk
Surrogate-based approval is not a shortcut without consequences when the underlying validation turns out to be wrong. The methodological literature on this point routinely cites the cholesteryl ester transfer protein (CETP) inhibitor torcetrapib as a cautionary example: it raised HDL cholesterol — the surrogate it was designed around — as intended, but its ILLUMINATE cardiovascular outcomes trial was halted in 2006 after the drug was associated with increased cardiovascular events and mortality rather than the expected benefit. The general lesson researchers and regulators draw from cases like this is that a surrogate’s correlation with an outcome under one mechanism of action does not guarantee the same relationship holds under a different mechanism, which is exactly why trial-level meta-analytic validation (rather than single-trial correlation) is now the preferred standard, and why accelerated approvals carry a mandatory confirmatory-trial obligation rather than standing on the surrogate result alone.
Frequently Asked Questions
What is the difference between a surrogate endpoint and a clinical endpoint?
A clinical endpoint (sometimes called a “true” or “hard” endpoint) directly measures how a patient feels, functions, or survives — overall survival, stroke, hospitalization. A surrogate endpoint is a different, typically earlier-measurable variable used as a substitute for that direct measure, valid only to the extent it has been shown to predict it.
Can a surrogate endpoint support full, traditional FDA approval?
Yes, if it meets the higher “validated surrogate endpoint” bar — strong mechanistic rationale plus sufficient clinical evidence that the relationship to clinical benefit is well established. A “reasonably likely” surrogate, by contrast, can only support accelerated approval, and comes with a mandatory postmarketing confirmatory trial requirement.
Do the Prentice criteria have to be fully met for a surrogate to be used in practice?
No. Full Prentice-style mediation is rarely demonstrable for any real biological pathway, so regulatory and methodological practice has largely moved to trial-level meta-analytic validation — testing whether treatment effects on the surrogate reliably predict treatment effects on the true outcome across multiple randomized trials — as the practical validation standard.
What happens if a surrogate endpoint used for accelerated approval turns out not to predict clinical benefit?
The sponsor is still obligated to complete the required postmarketing confirmatory trial. If that trial fails to verify the anticipated clinical benefit, or is not pursued with due diligence, FDA has authority to withdraw the accelerated approval through an expedited withdrawal process.
